A problem that occurs when trying to achieve nano-computing systems is to find system architectures that can be manufactured in practice. For example, manufacturing of nanoscale components is inherently unreliable and leads to large amounts of failed and poorly functioning devices. The architecture, the computing logic and the algorithms of such devices need to take this into account, and to be able to compensate for this unreliability somehow.
At present there are two basic options on how to solve this: try to improve hardware reliability by some means (e.g. via manufacturing process, redundant components), or to use software that can deal with unreliable hardware (e.g. via error correction, dynamic reconfiguration of hardware, robust algorithms or computing principles, like neural networks or any other machine learning algorithms capable of learning or being taught). Unfortunately, there is always a trade-off between hardware and software complexity: with simple software the hardware has to be very reliable, and with unreliable hardware the software required can be extensively complex.
One data processing device architecture that has been broadly used outside nano-scale implementation is the cross-bar architecture. Cross-bar structures have been proposed in U.S. Pat. No. 6,128,214 “Molecular wire cross-bar memory”, with certain types of binary state molecules in between, which can be tune on or off by applying voltage across the nanowires. Also, papers by Likharev cover similar ideas in devices that utilise neural networks. Furthermore, US20030236760 “Multi-layer training in a physical neural network formed using nanotechnology” and US2009/0043722 A1 “Adaptive neural network utilizing nanotechnology-based components” describe a neural-network type information processing devices utilizing nanoparticles. The devices described in this document consists of a cross-bar structure, with nanoparticles diluted in a solvent in between the top and bottom parts of the sandwiched structure.
WO 2009/013754 “Chemically sensitive field effect transistors and use thereof for electronic nose devices” describes how to use silicon nanowire field effect transistors (FETs) as chemical sensors, in conjunction with pattern recognition algorithms to detect the sensed chemicals. Here the pattern recognition algorithms are implemented in traditional CMOS devices
This same structure has been proposed for nano-implementation. The idea is that by crossing nanowires and functionalizing the crosspoint somehow with suitable molecules one can get a two-terminal transistor, which can be turned on and off by applying voltage to the wires. But there is still a problem that the system needs to interface at the microscale, so microscale connections to the nanowires are needed.
However, as stated above, this is difficult to do. Manufacturing can become complicated, and there are often losses due to poor contacts. In particular, the operation of the molecules in the cross-points is very limited, and only simple operations can be realized. There are also proposals of using cross-bar structures to perform Boolean computing (i.e. as FPGA processors) and as neural networks. With FPGA, the problem comes with the very large area required to construct even a simple circuit. The neural network-type systems suffer from the fact that (due to the cross-bar structure) the number of connections between the different logic gates or “neurons” are quite limited, which reduces the computing capability of the system. The number can be enhanced by adding microscale wires on top, but this complicates the manufacturing process. Due to the complications mentioned, no functioning cross-bar circuits have been demonstrated to date.
Another problem comes with effective analysis of sensory data. Nanoscale sensor elements have been fabricated from many materials, but the data needs to be analysed somehow to extract relevant features. Digital CMOS does not provide the best way of doing this, since it requires analog-to digital conversions, and can thus be very energy consuming compared to the energy used by the nanoscale sensor, particularly when the sensor data is very complex as its analysis can require a lot of processing.
The listing or discussion of a prior-published document or any background in this specification should not necessarily be taken as an acknowledgement that the document or background is part of the state of the art or is common general knowledge. One or more aspects/embodiments of the present disclosure may or may not address one or more of the background issues.